Single-Molecule SERS Discrimination of Proline from Hydroxyproline Assisted by a Deep Learning Model

Area of Science:

• Analytical Chemistry
• Biotechnology
• Artificial Intelligence in Medicine

Background:

• Detecting low-abundance hydroxylation is critical for early disease diagnostics and drug development.
• Existing single-molecule surface-enhanced Raman spectroscopy (SERS) methods face challenges with signal fluctuations and citrate interference.

Purpose of the Study:

• To develop a robust method for single-molecule discrimination of hydroxylation.
• To overcome signal interference and fluctuations in SERS analysis for hydroxylation detection.

Main Methods:

• Utilized a plasmonic particle-in-pore sensor for enhanced SERS signal acquisition.
• Employed an occurrence frequency histogram of single-molecule SERS spectra to extract spectral features.
• Applied a one-dimensional convolutional neural network (1D-CNN) model for hydroxylation discrimination.

Main Results:

• Successfully achieved single-molecule discrimination of proline and hydroxyproline using the 1D-CNN model with 96.6% accuracy.
• The histogram method effectively mitigated signal fluctuations and citrate interference, enabling clean signal generation for model training.
• Validated that the 1D-CNN model's extracted features directly corresponded to hydroxylation-induced spectral changes.

Conclusions:

• The integrated approach of particle-in-pore SERS, histograms, and 1D-CNN offers a powerful new tool for single-molecule hydroxylation detection.
• This method provides a significant advancement for early disease diagnostics and the development of targeted therapeutics.